Drafted with AI tools and edited by a human. The figures and conclusions were checked by a GPUniq editor.
TL;DR
OpenAI prices the GPT-Live 1.1 voice session API at $0.05 per minute. It packs a 128,000 token context window and scores 83.6% on first-attempt Tau3 task performance. This model improves on predecessors with reduced turn-taking latency and 2-hour maximum live session lengths.
What is GPT-Live 1.1?
OpenAI released GPT-Live 1.1 as an updated full-duplex voice model for ChatGPT Voice and API integrations. It handles native voice-to-voice workflows and natural conversation handling. Developers can run live conversation sessions up to 2 hours long per session. The model supports telephony and SIP out of the box, letting systems delegate reasoning and tool use to backend agents.
How much does the API cost?
Voice session pricing is set at $0.05 per minute. Developer billing uses a pay-as-you-go model based on active session duration.
Context Window Limits
The model features a 128,000 token context window. This large window allows applications to retain long-form conversational history during live calls. Older context gets summarized when capacity limits are crossed.
Tau3 Performance
GPT-Live 1.1 scores 83.6% on first-attempt Tau3 task performance.
| Model | Tau3 Task Performance, Percent | Turn-taking Latency, Seconds | Voice Options |
|---|---|---|---|
| GPT-Live 1.1 | 83.6% | 0.798 | 13 new voices and improved non-English accents |
| GPT-Realtime-2.1 | 45.7% | 1.41 | - |
Turn-Taking Latency
Turn-taking latency drops to approximately 0.798 seconds.
Conversational Benchmarks
The model scores 97.3% on standard Conversational Dynamics benchmarks. It handles interruptions and overlapping speech naturally. Multi-turn reasoning improves without losing conversational context, and the system easily manages short filler phrases like mhmm or yeah.
New Features and Voices
Thirteen new voices ship with this release, alongside expanded accent fluency for non-English languages like French, German, Hindi, Japanese, Korean, and Mexican Spanish. Developers can select specific voice options. Audio outputs include a SynthID watermark for provenance tracking.
Related on the GPUniq blog
Methodology
Facts are compiled from the vendor's own announcement and documentation, linked below, at the time of writing; where the vendor hasn't published a number it is marked as such. GPUniq prices, when the model is already available here, come from our live catalog.
Sources
- 1.Release Notes | OpenAI | Oct 15, 2026 — GPT-Live-1.1 rollout — OpenAI (accessed )
- 2.ChatGPT — Release Notes — Introducing GPT-Live-1 in ChatGPT Voice — OpenAI Help Center (accessed )
- 3.Build more natural voice experiences with GPT-Live-1 in the API — OpenAI — OpenAI (accessed )
Want cheap GPUs for your next project?
Browse live GPU prices and rent the right card in seconds — H100, A100, RTX 4090, and 50+ more models.
